A human-first approach to safe AGI
Hello, world.We are an indie AI lab based in the US, taking a human-first approach toward safe AGI. Most state-of-the-art AI models approximate reality statistically. Instead, we articulate our understanding of reality to our model directly.We develop symbolic systems with expressiveness in mind. This allows us to encode the nuances of the human soul, including the subtleties of taste shaped by lived experience.
Human first
道大，天大，地大，人亦大。The Tao is great. Heaven is great. Earth is great. Humanity is also great. — Tao Te Ching
For humanity to last, we must align ourselves with the Tao (道), for we are not separate from the forces that gave rise to us and continue to sustain us. Everything we do, including the technologies we create, should serve this alignment.This means staying true to the human spirit and clearing away what obscures it. Each of us has a unique way of seeing the world. Beauty emerges when we express that authentically, and with love. The realness in that expression also resonates with other souls. As AI automates more of our work, what remains for us to do is the lifelong practice of becoming more whole: cultivating our attention and deepening our capacity to love. This is a devotional path toward becoming one with the divine, the ultimate attainment.
How it worksAt the core of our work is a subject model: a structured representation of how a subject experiences the world. Implemented as a state machine, it makes the flow of of an agent's attention explicit and editable.This model serves as a vessel for imbuing our understanding of reality. Since no single perspective can capture the whole truth, we work with contributors who bring their own ways of seeing. Each perspective illuminates a different aspect of our shared reality.Together, we form a collective intelligence across domains that compounds over time.
DemoIn the example below, we show a simplified state machine that encodes learnings from timeless spiritual practices. These layers in the state machine help the agent sense and navigate:Try chatting with it. Is there anything on your mind that it can help unpack?
Absorption statesHow attention settles
Spatial mapWhere energy arises
EmotionsWhat named feelings are present
Cosmic guidanceGuidance for the next move
A · Spiritual state machine
gpt-6.1-sol
B · Regular LLM
gpt-6.1-sol
Your conversation will appear here.
Loading chat…
This is a minimal version for illustrative purposes. The agent is connected to no other knowledge context or tools.
Business modelTo sustain this work, our initial business model is to create education content. Using our own system, we are running a portfolio of brands that bring improvements to people's lives.As our system matures, we will continually open source parts of what we build. We will also start selling premium workflows and curated datasets that others build on.
Our team
What's nextThis is the beginning of a series of releases. In the coming weeks, we will be sharing more of our independent AI research, including our eval framework, other symbolic systems, and more.If our work resonates, we would love to stay in touch.